SPECIAL SECTION
BIOTROPICA 42(1): 41–48 2010
10.1111/j.1744-7429.2009.00569.x
An Untidy Cover: Invasion of Bracken Fern in the Shifting Cultivation Systems of
Southern Yucatán, Mexico
Laura C. Schneider1,3 and D. Nelun Fernando1,2
1
Department of Geography, Rutgers, The State University of New Jersey, 54 Joyce Kilmer Drive, Piscataway, NJ 08854, U.S.A.
2
Department of Geography, Faculty of Arts, University of Colombo, Reid Avenue, Colombo 3, Sri Lanka
ABSTRACT
In land change science studies, a cover type is defined by land surface attributes, specifically including the types of vegetation, topography and human structures, which
makes it difficult to characterize land cover as discrete classes. One of the challenges in characterizing a land-cover type is to distinguish variability within the class from
actual land-cover transformation. The spread of plant invasions in tropical systems is affected by seasonal variations and disturbances such as agricultural activities and
fires, making it difficult to determine the spread through thematic classifications. In this paper, we estimate the changes in spatial extent and seasonal variation of
bracken fern invasion in Southern Yucatán from 1989 to 2005 by using a linear mixture model (LMM), a widely used method in the classification of remotely sensed
data. The results show an increase in areas affected by bracken from 40 km2 in 1989 to almost 80 km2 in 2000. Lower estimates of the invasion resulted from data
acquired at the end of the dry season (March–May), when bracken mixes with secondary vegetation or is removed by fires. The accuracy of the maps is estimated
through the use of sketch maps of farmer’s parcels and field data collected from 2000 to 2001. Understanding the spatial distribution and annual variability of bracken
fern cover in the region is critical to determining the relation between disturbances such as fire and forest recovery. Using LMM may enhance this understanding by
giving a more accurate picture of the extent and distribution of bracken fern invasion.
Abstract in Spanish is available at http://www.blackwell-synergy.com/loi/btp
Key words: fires; invasive plants; land change science; linear mixture model; remote sensing.
INVASIVE PLANTS ARE SIGNIFICANT GLOBAL PHENOMENA that affect
biological diversity and ecosystem function (D’Antonio & Vitousek
1992, Mooney & Hobbs 2000, Lugo 2004, Lockwood et al. 2007).
Invasions occur naturally, but human activity often accelerates invasion rates by several orders of magnitude (Mack 2000, 2005).
Land-use studies related to invasive species have usually focused on
how land transformation, such as deforestation, drives the spread of
such species (Vitousek et al. 1997, Hobbs 2000). Recent studies
have shown how climate change, combined with disturbances from
human activity, will result in feedbacks that further favor the dispersion of invasive species (Vitousek et al. 1997, Mooney & Hobbs
2000, Crowl et al. 2008). Little attention, however, has been given
to the study of plant invasions as an important land-cover change
transition (Hobbs 2000), partly because of the difficulty of detecting and monitoring the spatial extent of such invasions (Schneider
2006, Turner et al. 2007).
To understand plant invasions, relate them to environmental
change, and predict future spread researchers have called for the
creation of a global information system that describes not only the
ecology and control methods of plant invasions, but also their spatial arrangement (Mack 2000, Richardson & Pyšek 2006, GISP
2008). With an increased interest in the use of earth observations
from satellites that provide repetitive and spatially explicit measurements of biophysical surface attributes such as vegetation cover, it
seems logical that the spatial arrangement of plant invasions should
be determined with the use of such tools (Lambin et al. 2006,
Bradley & Mustard, 2005, Schneider 2008). Yet, the spread of
Received 12 September 2008; revision accepted 16 May 2009.
3
Corresponding author; e-mail: [email protected]
plant invasions is often hard to characterize into discrete classes.
This limitation could result in merging land transformations caused
by the spread of invasive plants with more subtle changes that affect
a cover without changing the overall characteristics of the landscape
(e.g., seasonality). These challenges have motivated the development of continuous land-cover classes, a tool that allows the examination of the trajectory of complex land classes and the detection of
the effects of seasonality on vegetation cover (DeFries et al. 1999,
Roberts et al. 2002).
Spatial characterization of plant invasions in tropical landscapes using remotely sensed data is rare. Moreover, land change
science studies tend to focus on the thematic characterization of the
areas invaded, which has proved to be helpful for monitoring and
modeling the invasion (Bradley & Mustard 2005, Huang & Geiger
2008). Mapping plant invasion as a continuous class allows for a
better understanding of annual variations of the invasions, which is
critical in linking invasions to disturbance events such as fires and
land fragmentation. This type of analysis, which has been applied to
the study of other land transitions such as deforestation, has shown
that changes in land cover are not necessarily progressive and gradual but they exhibit complex trajectories such as periods of rapid
and abrupt change followed by a quick recovery (Archard et al.
2002, Lambin et al. 2006).
Remotely sensed data have proved to be fundamental in characterizing land-cover dynamics due to the spatial and temporal
quality of the data sets (Rindfuss et al. 2004, DeFries et al. 2005).
Such data sets strengthen studies in land change science; yet, these
tools have been less used by ecologists mainly due to the inability to
separate successfully invasive plants from other types of vegetation
through spectral signatures (Mack 2005). Also, most of the remote
r 2009 The Author(s)
Journal compilation r 2009 by The Association for Tropical Biology and Conservation
41
SPECIAL SECTION
42
Schneider and Fernando
sensed data come at resolutions that aggregate landscape attributes,
e.g., a 30 m pixel, could contain different types of vegetation and
could result in ambiguous characterizations. Some of these issues
can be resolved by using linear mixture model (LMM), a technique
commonly used in remote sensing when dealing with landscape
attributes smaller than the spatial resolution of the sensor. The assumption is that the sensor captures a signal that is the result of a
mixture of reflected radiance of the different material found in the
landscape, and so the value of the minimum spatial unit (pixel) is
the composite spectral signature of surface material such as vegetation, soils, shade and atmospheric effects (Sohn & McCoy 1997,
Roberts et al. 2002). The interpretation of the data then could be
improved by understanding the spectral components within each
pixel.
In this paper, we explore the utility of LMM analysis in determining the spatial configuration and variation of bracken fern invasion in the Southern Yucatán. Bracken fern invasion is an
important land transition in the region; over the past decade, areas
under bracken fern invasion have exceeded the areas under cultivation (Schneider 2006). Frequent fires and land clearance for agriculture have facilitated the replacement of secondary vegetation
with bracken fern, while competition between the invasive and native plants through succession has created mixed areas of invasive
species and secondary vegetation (Fig. 1). Areas of bracken fern
mixed with secondary vegetation have a different spectral response
(as detected by satellite sensors) from the areas where either bracken
fern or mature vegetation dominates (Roy Chowdhury & Schneider 2004, Schneider 2004). In thematic classifications, however,
mixed areas are usually considered either invaded or not, depending
on how close the value of reflectance of a pixel is to the mean value
for bracken fern cover. This could result in underestimating or
overestimating the invasion. Here, we rely on Landsat imagery to
examine the utility of using bracken fern and mature forest endmembers to determine the abundance and spatial extent of bracken
fern invasion in Southern Yucatán between 1989 and 2005. We
explore the interannual variability of bracken fern using data for
1995. Also, for a smaller area in the region, we used multispectral
high-resolution data analyzed through supervised classification for
2007 in order to estimate current areas of invasion and relate them
to previous patterns. We conclude with a discussion of the ecological implications of the trends of bracken invasion on the region.
METHODS
STUDY REGION.—The study region is located at the frontier of the
Mexican-Guatemalan border, an area of 3000 km2, bordering the
eastern buffer area of the Calakmul Biosphere Reserve (18127 0 N,
88150 0 W, 100–350 asl). The region is therefore considered a hotspot of tropical deforestation (Lepers et al. 2005; Fig. S1). The
southern Yucatán region contains the largest tract of continuous
deciduous forest in Mexico and it is experiencing an annual rate of
deforestation of 0.4 percent, comparable to other forest areas in
Central America (Perez-Salicrup 2004, Sader et al. 2004). The
study region is dominated by two types of seasonal tropical forests:
(1) semi-evergreen (semi-deciduous), well-drained upland forests of
medium stature with crown heights ranging from 15 to 20 m; and
(2) short-statured (5–10 m), less deciduous, seasonally inundated
forests (bajos) (Martı́nez & Galindo-Leal 2002, Vester et al. 2007).
Seasonally inundated savannas characterized by flood-tolerant forbes constitute another land-cover class that dominates to a lesser
extent. Other important land-cover classes related to human use are
secondary vegetation, mostly fallow, of 4–15 yr; agriculture, which
includes pastures, chili and swidden cultivated areas; and bracken
fern. Previous research shows that areas invaded by bracken were
areas cleared for agriculture and bracken-invaded areas are uncommon in areas surrounded by old growth forest (Schneider 2008).
Currently, land in the southern Yucatán is used mainly for agriculture, followed by pastures and forest extraction. The main activity of farmers is milpa cultivation—a polyculture system combining
maize, squash and beans. Households are also diversifying their
FIGURE 1. Conceptual model of interaction between land-use and bracken fern invasion.
SPECIAL SECTION
Invasion of Bracken Fern
incomes with off-farm activities. After cultivation, or during early
stages of fallow, grasses are planted on some land parcels for cattle
ranching. Land clearance either for cultivation or for pasture is one of
the main drivers of bracken invasion in the region (Schneider 2006).
BRACKEN FERN.—Pteridium aquilinum (L.) Kuhn has a cosmopolitan distribution (Marrs & Watt 2002). For regions of central and
South America a recent revision of the genus reports the presence of
P. aquilinum var caudatum (now P. caudatum) (Thomson 2005).
Bracken is considered an invasive species because of its tendency to
spread out of control, producing a monoculture that discourages
the growth of other plant species, thus posing a direct threat to biological diversity. Bracken fern establishes itself on areas dominated
by fires, deforestation and agricultural activities, causing severe
problems to both farmers and conservationists (Pakeman et al. 1996).
The biological adaptability of bracken fern to different environments makes it a successful invader (Pakeman & Marrs 1996,
Schneider 2008). Disturbances such as fire and land clearing promote the invasion, and land clearance, fire regimes and land management practices have contributed to the spread of bracken fern in
Southern Yucatán (Schneider 2006; Fig. 1). The increase of this
invasive could potentially lead to further land abandonment and
thus promote greater deforestation (Schneider & Geoghegan
2006). Specifying where bracken fern invades is therefore a critical
component of forest conservation and management in the region.
Land clearance has been found to be an important driver of
bracken fern invasion, but it is less understood how bracken fern
dominates over secondary growth. Possible explanations are that
bracken fern invasion competes successfully with secondary vegetation due to frequent fires and, perhaps, soil nutrient limitation
(Schneider 2004). Areas of bracken fern that are not continuously
burned present a thick ground layer of dry biomass and they can
support secondary woody vegetation. However, the added dry biomass makes infrequently burned patches more vulnerable to fires
(Fig. 1). These conditions appear to be met in bracken-secondary
vegetation parcels at least 3 yr in age. These parcels have enough
fuel to support fire ignition and bracken fern spread but are still too
young to support secondary trees that can suppress bracken. The
analysis of remote sensed data through a linear mixture analysis
provides the means to measure the fraction of bracken fern and
secondary growth in these invaded areas. Estimating such fractions
is critical for determining the possibility of areas invaded to burn
and/or recover from invasion.
LMM.—LMM is a common remote sensing technique used to determine the spectral composition of mixed areas (Foody et al. 1992,
Sohn & McCoy 1997, Roberts et al. 2002). The values of reflectance captured within a field of view of the sensor are the result of
the reflection of multiple materials that could be decomposed into
fractions of several unmixed spectra called end-members (Adams
et al. 1995). Unmixed spectra or pure cover types when mixed with
other covers could be measured by assuming a linear combination
of the values of reflectance of the ‘pure’ constituents in the different
spectral bands (Small 2004). Most of the studies that use LMM
analysis develop end-members or pure pixels that, combined, will
43
compose the areas of interest, mostly vegetation, nonphotosynthetic
vegetation and soils. For this study, we consider invaded bracken fern
areas, bright soils and old growth upland forest as end-members.
Lowland inundated forest and semi-inundated savannas are excluded
from the analysis because these types of forest are not affected
by bracken fern invasion; the inundation impedes the growth of
bracken fern populations (Schneider 2008). One of the goals of the
paper is to explore the utility of the linear mixture techniques when
looking at two spectral, separable vegetated covers.
The characterization of bracken fern fractions to determine the
spread is based on anniversary dates of Thematic Mapper (TM)LANDSAT 5 and Enhanced Thematic Mapper-LANDSAT 7
imagery. The imagery was acquired during the period of late
November to late January, which corresponds with the beginning
of the dry season, few months before fire events are used to prepare
land for agriculture. The specific dates are 17 November 1989, 26
November 1995 and 24 January 2000. For 2005, we used data for
19 April, which is the middle of crop establishment. The scene for
2005 is a product developed by USGS—SLC-off gap filling processing which corrects the stripping caused by the malfunction of
the sensor after 2003 (Scaramuzza et al. 2004). The LMM results
for 2005 are affected by the quality of the data set; but they provide
a general representation of areas affected by fires and bracken fern.
To determine intra-annual variability we analyzed data for 15
March 1995 and compared it with the November image of the
same year. All the data are available on the United States Geological
Survey’s the Glovis visualization viewer (http://glovis.usgs.gov).
The individual scenes were originally geo-referenced to a UTM
projection (UTM 16, Datum WGS84, Mexico) to under 0.5 pixel
RMS error. The scenes were corrected radiometrically by transforming digital numbers to values of reflectance following Chander
and Markham (2003). The reflectance bands used in the analysis
are those from the visible part of the spectrum (blue, green and red),
near infrared (NIR) and two short-wavelength infrared (SWIR-5
and SWIR-7).
End-members were chosen for bracken fern areas, old growth
upland forest and bright soils (limestone pits) (Figs. 2 and S2). Pure
pixels representing bracken fern, mature upland forest and bright
soils were selected from the center of homogeneous plots of these
land-cover types. In order to accommodate the dimensionality of
the mixing space we include bright soils as an end-member (Small
2004). This implies that bright soils are quite different and unique
spectrally from other covers in the region. Between 10 and 15
training sites of ‘pure’ end-members were defined using on-screen
site seeding and compared for consistency. The end-members were
located in the same area for all the dates except for areas of soils that
correspond to areas where pits for road construction are built. In
addition, only few training sites of bright soils were used because
they are rare in the region. The training sites used to develop landcover class end-members come from direct observations in the field
obtained in 2000–2001 and 2005 where the locations were recorded using Global Positioning Systems (GPS). Data on land
cover for the years before fieldwork were collected through detailed
sketch maps of land-use history developed with the help of local
farmers (Schneider 2006). All ground truth data were collected
SPECIAL SECTION
44
Schneider and Fernando
FIGURE 2. End-member spectral signatures from (A) 1989, (B) 1995, and
(C) 2000.
through field visits between 2000 and 2001. Topographic and thematic vegetation maps provided additional information on features
such as water bodies, inundated savannas and large seasonally
inundated bajos.
Fractions of the three end-members were estimated using
LMM. Specifically, we used the UNMIX command in the GIS
software IDRISI-Andes. The program assumes that a pixel value is a
combination of the means of the signatures of all the classes present
in the pixel—in this case the signatures of the end-members.
The calculations are described in Appendix S1. The input for the
calculations is the multispectral preprocessed scenes, in our case
windowed to the area of interest (36 90 km) and the signatures
from end-members. Areas of bajos, inundated savannas and water
bodies determined by thematic classification were excluded from
the area of analysis. The output of UNMIX includes an image for
each class defined by the three end-members indicating the percent
of that cover in each pixel and a residual image indicating how well
the actual pixel values match the calculated mixture of classes (Fig.
S3). The higher the residual value, the less likely the calculated
mixture is the mixture that actually exists.
The LMM results for 2005 are affected by the quality of the
data set, which resulted in misrepresenting the fractions obtained.
To determine areas affected by the invasion, we developed a supervised classification for this scene to determine the areas affected by
fires that were previously invaded. Then we added the estimation to
the final results.
To determine the accuracy of the bracken fern fractions we
used field data points collected during fieldwork in 2000 and 2001.
For accuracy of 1989 data, we used information from sketch maps
developed with the help of farmers. Thirty-one farmers were interviewed and the parcels were geo-referenced with the use of GPS.
During each interview, the parcel was walked through and reference
points were taken around bracken areas, cultivated plots and
boundaries of the parcel. Farmers were asked about the history of
bracken fern areas since the first time the parcels were used for
agriculture. The locations with bracken presence were used for accuracy assessment. The Kappa Index was calculated for each year
based on presence/absence of bracken fern areas in sketch maps.
A total of 104 points from the visits were used. From the LMM
results, fractions of 50 percent or more were considered to be invaded.
Finally, to evaluate the most recent levels of invasion in the
region and to estimate the general trends in the spread we used a
multispectral high-resolution quick-bird (3 mts) data set for April
2007. ETM1 cannot be used more recently due to the problem
with the stripping errors mentioned before. In the high-resolution
data set, the areas affected by bracken are visually and spectrally
distinctive from areas of secondary growth and agriculture. The
maps developed were also verified by a field visit in January 2008.
Areas of bracken were classified using supervised classification. The
resulting bracken area mapped from quick-bird was masked and
overlaid to the different LMM bracken fern fractions to determine
the trend of the invasion (Fig. 3).
RESULTS
CHANGE IN COVER 1989–2005.—Regional land-cover maps were
created to estimate bracken fern fractions from 1989 to 2005 (Fig.
S3A). The region overall is experiencing an increase (40 km2 until
2000) in areas affected by invasion, although a decrease is observed
in 2005 (Fig. 3). To tease apart the effect of seasonality and leaf
phenology in the final estimations, the LMM analysis is done on
data sets that correspond to anniversary dates. The Landsat TM-5
scenes were acquired from late November to late January, which
corresponds to the beginning of the dry season and few months
after maize cultivation. For 2005 the data used correspond to
SPECIAL SECTION
Invasion of Bracken Fern
FIGURE 3. Total land affected by bracken fern from 1989 to 2005 from
LMM results.
April 2005, at the beginning of land preparation for cultivation, the
time when fires are more frequent (J. Rogan et al. unpubl. data).
Unfortunately, a high-quality anniversary date scene for 2005 is not
available.
In general, the spectral signature of bracken is quite distinct
from other types of vegetation in the region (Fig. 2). The signatures
for both upland forest and bracken fern follow the same pattern:
lower values of reflectance in the red part of the spectrum and high
values on the NIR and SWIR. The major difference is observed in
higher values of reflectance in NIR for bracken. The value of separability using the transformed divergence index (Richards & Jia
1999) between bracken fern training sites and upland forest was
never below 1800. The values of separability range from 0 to 2000
and a value of 1500 or below is considered as a low separability. In
general, the larger the magnitude of the transformed divergence index, the greater the statistical distance between training sites and the
higher the probability of correct classification (Richards & Jia
1999). Of an average of 12–15 training sites for bracken fern and
upland forest used per image, no one has a value of separability
o 1800.
The residual maps (Fig. S3B) show that areas with higher values correspond to inundated savannas that were not removed from
the initial analysis and areas under cultivation. Higher values indicate that the end-members used in the analysis represent poorly the
spectral characteristics of those pixels.
INTRA-ANNUAL VARIATION IN BRACKEN COVER.—A large variation in
the amount of area invaded by bracken could be observed between
March 1995 and November 1995. The fractions estimated for
bracken fern in March 1995 were lower than those in the November scene. For March 1995, some of the invaded areas estimated in
November 1995 are burned and areas with higher bracken fractions
in November appeared to represent higher fractions of primary
vegetation in March. The analysis is only done for 1995 because of
the lack of cloud free scenes during the same period for 1989 and
2000.
There is a variation in the fractions and spatial distribution of
bracken fern areas within a year. Using a smaller area fitting the
45
quick-bird scene we see a general trend of increase in the total of
areas invaded by bracken from 1989 to 2007 (Fig. 3). The fractions,
however, are quite different between March and November of
1995. Areas burned in March are not considered to be invaded, but
in reality those areas usually get invaded quickly after fires (Fig. S4).
Multispectral high-resolution data for 2007 help us illustrate trend
and intra-annual variations of bracken fern cover. The data provide
a recent point in time which accurately represents spatial distribution of bracken and which we cannot achieve using recent Landsat
data sets.
Table 1 shows the values of accuracy of the maps produced for
this study. Kappa Index of Agreement (KIA) was calculated for
1989 and 2000. It is usually quite challenging to assess the accuracy
of fraction historical land-cover maps due to the lack of ground
truth data. Hence, from the historical information gathered
through the sketch maps we see that the overall KIA is 0.41. The
error of omission, i.e., areas of bracken that were classified as not
bracken, is 38 percent, however. For 2000, the error of omission is
lower, only 30 percent, and the overall KIA is higher (0.53) than for
1989. The overall KIA for both maps are low compared with other
classification analysis. Such low values could be explained in part by
the small percentage of total area under bracken—2 percent of total
area in the region (Pontius & Schneider 2001). Accuracy results of
other LMM studies discuss the difficulty of getting the appropriate
ground data and the effect of phenology on the data used for the
analysis (Sohn & McCoy 1997, Roberts et al. 2002).
DISCUSSION
Characterizing tropical land covers at a regional level proves to be
challenging as land covers in tropical areas are untidy and land
transformations are quite dynamic. Unlike hard classifiers such as
TABLE 1. Kappa Index of Agreement (KIA) for LMM bracken fractions for 1989
and 2000. Numbers in bold are the points where classification and field
points are in agreement.
Ground
Ground
truth
truth
bracken
nonbracken
Error
Total
commission
1989
Bracken fraction
34
1
35
0.0285
Nonbracken fraction
21
48
69
0.3043
Total
55
49
104
Error of omission
0.3818
KIA for bracken
0.4079
0.0204
Overall error: 0.2115
2000
Bracken fraction
Nonbracken fraction
49
21
9
25
58
46
Total
75
34
104
Error omission
0.30
KIA for bracken
0.5254
0.1552
0.4565
0.26
Overall error: 0.2885
SPECIAL SECTION
46
Schneider and Fernando
maximum likelihood that are commonly used to analyze remotely
sensed data, soft classifiers such as LMM defer making a unique
decision in terms of class membership of any pixel in favor of producing a group of statements about the degree of membership of
that pixel in each of the possible classes or end-members (Foody
& Cox 1994, Eastman et al. 2002). LMM has been commonly used
to characterize Landsat 30 m pixels as a combination of very distinct
components in the landscape, such as green vegetation, nonphotosynthetic vegetation and bright and dark soils. LMM has proved to
be important for local specific studies in discriminating vegetation
covers from desert shrubland (Sohn & McCoy 1997) and different
types of secondary vegetation in tropical forests of Amazonia (Roberts et al. 2002) as well as in determining the abundance of generic
attributes, which are not necessarily site specific (Small 2004).
Using two types of green vegetation as end-members in LMM, like
we do in this study, has not been explored in dense vegetated
areas mainly due to the difficulty of finding consistently spectrally
pure pixels that have enough difference to be considered endmembers.
We attempted to use bracken fern and mature upland forest as
two separated end-members because of the presence of large areas of
bracken in the region, the spectral separability from other landcover classes in the region, and the consistency of spectral signature
for different years and permanence of patches through time (Fig.
S2). The major difference in the spectral response of homogeneous
areas of bracken is the higher values of reflectance in the NIR waveband during the beginning of the dry season. Mature upland forest
in this region is characterized by high presence of semi-deciduous
species. A possible explanation in the difference of values of NIR
could be that by the time bracken fern areas are quite established
with several layers of individuals and their large leaf extent, they
begin to contrast with trees that have smaller leaves and that are
beginning to senesce and die at the end of the season. At the end of
the dry season, many areas of bracken are burned and bracken fern
individuals start to dry up resulting in a spectral response that reflects more areas under agriculture, spectrally showing nonphotosynthetic material and exposed dark soils. To determine bracken
fern spread through time the best dates for imagery to be analyzed is
during the beginning of the dry season, between November and
January.
Research results demonstrate how remotely sensed data analysis is helpful in determining the current extent of bracken fern in
the region. Bracken fern displays a unique spectral signature and
profile detected by TM and ETM1 sensors. Bracken fern also has a
high value of spectral separability from other land-cover classes in
the region. Pure stands of the fern larger than a pixel are readily
identifiable in the imagery; smaller-sized stands and stands that become mixed with shrubby growth are more difficult to separate
from a few other land covers. LMM is useful in determining fractions of bracken within a pixel. Such information is critical to the
prediction of future spread and the effects of the spread on fire dynamics in the region (J. Rogan et al. unpubl. data). Areas of bracken
fern that are not continuously burned present a thick ground-layer
of dry biomass and harbor secondary woody vegetation. The added
fuel makes infrequently burned patches more vulnerable to fires. In
this case, bracken fern fields exhibit more of a continuum than a
well-defined class.
Discrete classes are useful for models predicting the spread of
plant invasions as it provides both spatial and tabular measurements
to use in common modeling tools (Schneider & Geoghegan 2006).
The limitation with discrete classifications arises when the objective
of characterizing the spatial patterns is not just predicting but understanding the relations of the invasive with the natural vegetation
and land management practices. Understanding these relations is
important because it could elucidate the traits that make this invasive so successful in the region.
The results obtained from previous supervised classification
analysis (Schneider 2008) show how dense bracken fern areas are
distinctive in their spectral signature and relatively easy to identity
when well established and frequently burned, because fire removes
competitors and creates a homogeneous cover of the fern. Confusion develops where the fern escapes burning for more than 2 yr. In
such cases, competitors emerge and complicate the spectral signature of vegetation. LMM analysis is a helpful tool to determine the
degree of which bracken fern is mixed with secondary vegetation.
Through LMM analysis of a larger time series, we should be able to
estimate fire susceptibility of bracken fern fields based on time of
establishment.
Based on this study, we think a thematic classification tends to
overestimate the area under bracken, forcing some mixed pixels
to be considered 100 percent pure bracken. An interesting result to
point out is that for March of 1995 a decrease of bracken area is
observed, explained mostly by competition with secondary growth.
In November of 1995, however, we observe a stepped increase of
bracken again. The lower estimate in March 1995 does not imply a
decrease in the invasion, but that bracken was suppressed by secondary forest. Fires usually removed such competition from bracken, but due to bracken’s vegetative reproductive strategy through
rhizomes, areas cleared then were easily and quickly covered again
by bracken.
Remote sensing data are helpful to ecologists looking at the
spatial patterns of plant invasions. Still there is a need to expand
further the study of ecological processes linked to remotely sensed
data, which would result in a more accurate understanding of plant
invasion and its relation to disturbances. For example, the fractional
results of an LMM—ranging from low to high pixel proportions—
could be linked to fire frequencies. Preliminary results using fire
product from MODIS data show that between 2000 and 2006 areas cover by bracken have the largest fire frequency, followed by
early secondary growth areas (J. Rogan et al. unpubl. data). The
next step is to determine how mixed invaded areas relate to fire frequencies. If bracken fern areas are more mixed, that could indicate
that such areas are more likely to burn than pure bracken areas.
CONCLUSIONS
The general assumption explaining the spread of plant invasions is
that land degradation and increases in disturbance promote their
spread (Vitousek et al. 1997, D’Antonio & Kark 2002). The spatial
distribution of plant invasions and the relation with land use,
SPECIAL SECTION
Invasion of Bracken Fern
however, may suggest a more complex process. The results in this
paper illustrate how remote sensing analysis provides ways to characterize the spatial configuration of plant invasions and to detect
important intra-annual variations that relate to the way the invasive
relates to the local vegetation.
Most of the research on plant invasions originates in the biological sciences. Those studies focus on understanding the ecological dynamics that promote the invasions, and identify ways of
controlling the invasion through land management practices (Higgins et al. 1999). Bracken fern has become an important component of the coupled human–environment system of the Southern
Yucatán Peninsular Region. Here, secondary vegetation growth is
usually rapid with an increase in primary productivity
for areas that have been used for agriculture (Read & Lawrence
2003). Even if land uses such as agriculture promote great disturbance, vegetation may still move in and occupy the ground in relatively short periods of time and succession typically proceeds
quickly. The ability of the vegetation to reclaim quickly cleared
ground allows for the recovery of nutrients in the soil, which enables the success of shifting cultivation, usually involving a long
fallow period after a brief interval of cultivation (Turner & Geoghegan 2003). Such process of succession, however, is affected by the
ability of bracken fern to dominate after land clearance and
fire (Fig. 1).
Previous research considered land use as one of the primary
drivers of the increase in plant invasions. Accordingly, socioeconomic factors such as the availability of labor and capital and the
dependence of households on subsistence practices determined the
willingness or ability of farmers to combat bracken fern invasion
(Schneider 2006, 2008). The patterns of plant invasions presented
in this paper suggest a more complex relation with disturbances
such as fire than being entirely driven by land management decisions. This illustrates how the ‘generalities’ about land-cover change
deriving from a rich array of regional case studies may improve the
understanding and modeling of critical themes in global change
and sustainability studies.
ACKNOWLEDGMENTS
Research was supported by NASA’s Land-Use and Land-Cover
Change program through grant NAG 511134; the Gordon and
Betty Moore Foundation; and Rutgers Research Council Grant.
We would like to thank M. Uriarte, M. Labban and two anonymous reviewers for comments on early versions of the manuscript.
SUPPORTING INFORMATION
Additional Supporting Information may be found in the online
version of this article:
FIGURE S1. Southern Yucatán Peninsular Region. Study Region.
FIGURE S2. (A) Landsat TM-5 composites of Red, NIR and
SWIR for 1989 and 2000 showing areas invaded by bracken fern.
47
(B) Close up of areas invaded with bracken fern in eastern Yucatan
region from March to November, end of cropping cycle.
FIGURE S3. (A) Bracken Fern Fractions using Linear Mixture
Model from 1989 to 2000. (B) Residuals image indicating how well
the actual pixel values match the calculated mixture of classes.
FIGURE S4. Changes in bracken fern fractions from March to
November in 1995.
APPENDIX S1. Additional methods.
Please note: Wiley-Blackwell are not responsible for the content or
functionality of any supporting materials supplied by the authors.
Any queries (other than missing material) should be directed to the
corresponding author for the article.
LITERATURE CITED
ADAMS, J. B., D. E. SABOL, V. KAPOS, D. A. ROBERTS, AND M. O. SMITH. 1995.
Classification of multispectral images based on fractions of endmembers:
Application to land-cover change in the Brazilian Amazon. Remote
Sens. Environ. 52: 137–154.
ARCHARD, F. H., D. EVA, H. STIBIG, P. MAYAUX, F. GALEGO, T. RICHARDS, AND
J. MALINGREAU. 2002. Determinations of deforestation rates of the
world’s humid forest tropical forest. Science 297: 999–1002.
BRADLEY, B. A., AND J. F. MUSTARD. 2005. Identifying land cover variability distinct from land cover change: Cheatgrass in the Great Basin. Remote
Sens. Environ. 94: 204–213.
CHANDER, G., AND B. MARKHAM. 2003. Revised Landsat-5 TM radiometric calibration procedures and postcalibration dynamic ranges. IEEE Trans.
Geosci. Remote Sens. 41: 2674–2677.
CROWL, T., T. CRIST, R. R. PARMENTER, G. BELOVSKY, AND A. E. LUGO. 2008.
The spread of invasive species and infectious disease as drivers of ecosystem change. Front. Ecol. 6: 238–246.
D’ANTONIO, C. M., AND S. KARK. 2002. Impacts and extent of biotic invasions in
terrestrial ecosystem. Trends Ecol. Evol. 17: 202–204.
D’ANTONIO, C., AND P. VITOUSEK. 1992. Biological invasions by exotic grasses,
the grass-fire cycle and global change. Annu. Rev. Ecol. Syst. 23: 63–88.
DEFRIES, R., G. P. ASNER, AND R. HOUGHTON. 2005. Trade-offs in land-use
decisions: Towards a framework for assessing multiple ecosystem
responses to land-use change. In R. DeFries, G. P. Asner, and R.
Houghton (Eds.). Ecosystems and land use change, pp. 1–9. American
Geophysical Union, Washington, DC.
DEFRIES, R. S., J. R. G. TOWNSHEND, AND M. C. HANSEN. 1999. Continuous
fields of vegetation characteristics at the global scale at 1-km resolution.
J. Geophys. Res. 104: 16911–16923.
EASTMAN, J. R., H. JIANG, H. ZHU, S. CREMA, AND J. TOLEDANO. 2002 In-process
classification assessment (IPCA) using a Dempster-Shafer decomposition of classification uncertainty. 29th International Symposium on
Remote Sensing of the Environment, Buenos Aires, Argentina.
FOODY, G. M., N. A. CAMPBELL, N. M. TRODD, AND T. F. WOOD. 1992. Derivation and applications of probabilistic measures of class membership
from the maximum-likelihood classification. Photogram. Eng. Remote
Sens. 58: 1335–1341.
FOODY, G. M., AND D. P. COX. 1994. Sub-pixel land cover composition estimation using a linear mixture model and fuzzy membership functions. Int.
J. Remote Sens. 15: 619–631.
GISP. 2008. The Global Invasives Program. Available at http://www.gisp.org/
(accessed August 30, 2008).
HIGGINS, S. I., D. M. RICHARDSON, R. M. COWLING, AND T. H. TRINDER-SMITH.
1999. Predicting the landscape-scale distribution of alien plants and
their threat to plant diversity. Conserv. Biol. 13: 303–313.
SPECIAL SECTION
48
Schneider and Fernando
HOBBS, R. J. 2000. Land-use changes and invasions. In H. A. Mooney and R. J.
Hobbs (Eds.). Invasive species in a changing world, pp. 55–65. Island
Press, Washington, DC.
HUANG, C., AND E. GEIGER. 2008. Climate anomalies provide opportunities for
large-scale mapping of non-native plant abundance in desert grasslands.
Divers. Distrib. 14: 875–884.
LAMBIN, E. F., H. GEIST, AND R. RINDFUSS. 2006. Local processes with global
impacts. In E. F. Lambin and H. J. Geist (Eds.). Land-use and landcover change: Local processes and global impacts, pp. 1–8. Springer,
Berlin, Heidelberg, Germany.
LEPERS, E., E. F. LAMBIN, A. C. JANETOS, R. DEFRIES, F. ACHARD, N. RAMANKUTTY, AND R. J. SCHOLES. 2005. A synthesis of information on rapid
land-cover change for the period 1981-2000. Bioscience 55: 115–124.
LOCKWOOD, J., M. HOOPES, AND M. MARCHETTI. 2007. Invasion ecology. Blackwell Publishing, Malden, Massachusetts.
LUGO, A. E. 2004. The outcome of alien tree invasions in Puerto Rico. Front.
Ecol. Environ. 2: 265–273.
MACK, R. N. 2000. Assessing the extent, status and dynamism of plant invasions:
Current and emerging approaches. In H. A. Mooney and R. J. Hobbs
(Eds.). Invasive species in a changing world, pp. 141–168. Island Press,
Washington, DC.
MACK, R. N. 2005. Assessing biotic invasions in time and space: The second
imperative. In H. A. Mooney, R. N. Mack, J. McNeely, L. Neville, P. J.
Schei, and J. Waage (Eds.). Invasive alien species: A new synthesis, pp:
179–208. Island Press, Washington, DC.
MARRS, R. H., AND A. S. WATT. 2002. Biological Flora of the British Isles:
Pteridium aquilinum (L.) Kuhn. J. Appl. Ecol. 94: 1272–1321.
MARTÍNEZ, E., AND C. GALINDO-LEAL. 2002. La vegetación de Calakmul:
descripción, composición y distribución. Bol. Soci. Bot. Méx. 71: 7–32.
MOONEY, H. A., AND R. J. HOBBS. 2000. Invasive species in a changing world.
Island Press, Washington, DC.
PAKEMAN, R. J., AND R. H. MARRS. 1996. Modelling the effects of climate change
on the growth of bracken (Pteridium aquilinum) in Britain. J. Appl.
Ecol. 33: 561–575.
PAKEMAN, R. J., R. H. MARRS, D. C. HOWARD, C. J. BARR, AND R. M. MULLER.
1996. The bracken problem in Great Britain: Its present extent and
future changes. Appl. Geogr. 16: 65–86.
PEREZ-SALICRUP, D. 2004. Forest types and their implications. In B. L. Turner
II, J. Geoghegan, and D. Foster (Eds.). Integrated land-change science
and tropical deforestation in the Southern Yucatán: final frontiers, pp.
63–80. Clarendon Press of Oxford University Press, Oxford, UK.
PONTIUS, R. G. JR., AND L. C. SCHNEIDER. 2001. Land-use change model validation by a ROC method for the Ipswich watershed, Massachusetts,
USA. Agric. Ecosyst. Environ. 85: 239–248.
READ, L., AND D. LAWRENCE. 2003. Recovery of biomass following shifting
cultivation in dry tropical forests of the Yucatan. Ecol. Appl. 13: 85–97.
RICHARDS, J. A., AND X. JIA. 1999. Remote sensing digital image analysis (3rd
Edition). Springer-Verlag, Berlin, Germany.
RICHARDSON, D. M., AND P. PYŠEK. 2006. Plant Invasions: Merging the concepts
of species invasiveness and community invasibility. Prog. Phys. Geogr.
30: 409–431.
RINDFUSS, R. R., S. WALSH, B. L. TURNER II, J. FOX, AND V. MISHRA. 2004.
Developing a science of land change: Challenges and methodological
issues. Proc. Natl. Acad. Sci. 101: 13976–13981.
ROBERTS, D. A., I. NUMATA, K. HOLMES, G. BATISTA, T. KRUG, A. MONTEIRO, B.
POWELL, AND O. A. CHADWICK. 2002. Large area mapping of land-cover
change in Rondonia using multitemporal spectral mixture analysis and
decision tree classifiers. J. Geogphys. Res. 107 (D20): 40-1–40-16.
ROY CHOWDHURY, R., AND L. C. SCHNEIDER. 2004. Land-cover/use in the southern Yucatán peninsular region, Mexico: Classification and change analysis. In B. L. Turner II, J. Geoghegan, and D. Foster (Eds.). Integrated
land-change science and tropical deforestation in the Southern Yucatán:
final frontiers, pp: 105–141. Clarendon Press of Oxford University
Press, Oxford, UK.
SADER, S. S., R. ROY CHOWDHURY, L. C. SCHNEIDER, AND B. L. TURNER II.
2004. Forest change and human driving forces in Central America.
In G. Gutman, A. Janetos, C. O. Justice, E. Moran, J. Mustard, R.
Rindfuss, D. Skole, B. L. Turner, and M. Cochrane (Eds.). Land change
science: Monitoring, and understanding trajectories of change on
earth’s surface, pp. 57–76. Kluwer Academic Publishers, Dordrecht,
The Netherlands.
SCARAMUZZA, P., E. MICIJEVIC, AND G. CHANDER. 2004 Landsat-7 ETM1
SLC Gap-filled products methodology. Final report to USGS, EROS
Data Center, March 2004. Available at http://landsat.usgs.gov/
documents/SLC_Gap_Fill_Methodology.pdf (accessed January 15,
2009).
SCHNEIDER, L. C. 2004. Bracken Fern (Pteridium aquilinum (L.) Kuhn) invasion
in Southern Yucatán peninsular region: A case for land-change science.
Geogr. Rev. 94: 229–241.
SCHNEIDER, L. C. 2006. The effect of land management practices on bracken
fern invasion in the region of Calakmul, Mexico. J. Latin Am. Geogr. 5:
91–107.
SCHNEIDER, L. C. 2008. Plant invasions in an agricultural frontier: Linking satellite, ecological and household survey data. In A. Millington and W.
Jepson (Eds.). Land change science in the tropics: Changing agricultural
landscapes, pp. 117–142. Springer, New York, New York.
SCHNEIDER, L. C., AND J. GEOGHEGAN. 2006. Land abandonment in an agricultural frontier after a plant invasion: The case of Bracken Fern in Southern Yucatán, Mexico. Agric. Resour. Econ. Rev. 35: 167–177.
SMALL, C. 2004. The Landsat ETM1 spectral mixing space. Remote Sens.
Environ. 93: 1–17.
SOHN, Y., AND R. M. MCCOY. 1997. Mapping desert shrub rangeland using
spectral unmixing and modeling spectral mixtures with TM data. Photogram. Eng. Remote Sens. 63: 707–716.
THOMSON, J. A. 2005. Taxonomy and evolutionary relationships in bracken: A
worldwide perspective on the genus pteridium for 2005. International
Bracken Group, Liverpool, UK.
TURNER, B. L. II, AND J. GEOGHEGAN. 2003. Land-cover and land-use change
(LCLUC) in the southern Yucatán peninsular region (SYPR): An Integrated approach. In J. Fox, R. R. Rindfuss, S. J. Walsh, and V. Mishra
(Eds.). People and the environment: Approaches for linking household
and community surveys to remote sensing, pp. 31–60. Kluwer Academic
Publishers, Boston, Massachusetts.
TURNER, B. L. II, E. LAMBIN, AND A. REENBERG. 2007. The emergence of land
change science for global environmental change and sustainability. Proc.
Natl. Acad. Sci. USA 104: 20666–20671.
VESTER, H. M., D. LAWRENCE, J. R. EASTMAN, B. L. TURNER, S. CALMÉ, R.
DICKSON, C. POZO, AND F. SANGERMANO. 2007. Land change in the
Southern Yucatán and Calakmul biosphere reserve: Effects on habitat
and biodiversity. Ecol. Appl. 17: 989–1003.
VITOUSEK, P. M., H. MOONEY, J. LUBCHENCO, AND J. MELILLO. 1997. Human
domination of earth’s ecosystems. Science 277: 494–500.